Centering the Margins: Outlier-Based Identification of Harmed Populations in Toxicity Detection

Vyoma Raman, Eve Fleisig, Dan Klein

Submission Type: Regular Long Paper

Submission Track: Ethics in NLP

Submission Track 2: Computational Social Science and Cultural Analytics

Keywords: AI fairness, AI ethics, toxicity detection, hate speech detection, outlier analysis, marginalization, harm measurement

TL;DR: We operationalize the notion of social marginalization through outlier detection to uncover patterns of harm faced by certain demographic groups, demonstrating that model performance is consistently and significantly worse for demographic outliers.

Abstract: The impact of AI models on marginalized communities has traditionally been measured by identifying performance differences between specified demographic subgroups. Though this approach aims to center vulnerable groups, it risks obscuring patterns of harm faced by intersectional subgroups or shared across multiple groups. To address this, we draw on theories of marginalization from disability studies and related disciplines, which state that people farther from the norm face greater adversity, to consider the "margins" in the domain of toxicity detection. We operationalize the "margins" of a dataset by employing outlier detection to identify text about people with demographic attributes distant from the "norm". We find that model performance is consistently worse for demographic outliers, with mean squared error (MSE) between outliers and non-outliers up to 70.4% worse across toxicity types. It is also worse for text outliers, with a MSE up to 68.4% higher for outliers than non-outliers. We also find text and demographic outliers to be particularly susceptible to errors in the classification of severe toxicity and identity attacks. Compared to analysis of disparities using traditional demographic breakdowns, we find that our outlier analysis frequently surfaces greater harms faced by a larger, more intersectional group, which suggests that outlier analysis is particularly beneficial for identifying harms against those groups.

Submission Number: 5132